JP2924691B2 - Quantization noise reduction method and image data decoding device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for reducing quantization noise generated when decoding transform-coded image data, and decoding image data for decoding transform-coded image data by applying the above method. Device.

[0002]

2. Description of the Related Art In the case of transmitting, recording, and reproducing image signals, audio signals, and other various signals as digital signals,
An information compression / decompression technique is used. In other words, for example, when digitizing an image signal or an audio signal, transmission is performed only by performing linear quantization (equal quantization) in which each sample value is replaced with one representative value of signal levels obtained by equally dividing each sample value. , The information amount of the signal to be recorded and reproduced is
This is because the state becomes very large. Conventionally, in the technical field of broadcast communication and the technical field of recording / reproducing, for example, human vision and hearing are sensitive in portions where signal changes are small, and there are some errors in portions where signal changes are severe. However, in addition to utilizing the characteristics of human vision and hearing that are difficult to detect to reduce the amount of information per sample, the application of many information compression technologies has enabled transmission, recording and playback. It is well known that high-efficiency compression technology (high-efficiency information encoding technology) for various types of information to be targeted has been put into practical use.

[0003] The amount of information per hour in a moving image of a quality similar to that of a reproduced image displayed by using a reproduced signal from a VHS (registered trademark) VTR currently in practical use is approximately 109 Gbits. In addition, the amount of information per hour in a moving image having an image quality equivalent to that of a received image of the current standard color television system in Japan is approximately 360 Gbits, but has a large amount of information as described above. Practical research on high-efficiency compression of image information required to transmit, record, and reproduce image information using current transmission lines and recording media that have been put into practical use has been actively conducted. .

In the high efficiency compression method of image information, which is currently proposed as a high efficiency compression method of image information, there is a high correlation between adjacent pixels in a natural image. The amount of information to be compressed using the correlation (compression of the amount of information performed by using spatial correlation) and the amount of information to be performed by using the correlation between screens (between frames) arranged on the time axis The amount of information is compressed by combining three different types of compression means: compression (compression of information using temporal correlation) and compression of information due to bias in code appearance probability. Efficiency coding is achieved. Many methods have been proposed as means for compressing the information amount of an image using the above-mentioned correlation within a screen (in a frame). Recently, however, KL ( Karhunen-Loeve transform, discrete cosine transform (DCT), discrete Fourier transform,
Orthogonal transforms such as Walsh-Hadamard transform and the like are often used.

[0005] For example, MPEG (Moving Picture C) established under the ISO (International Organization for Standardization).
The coding system (also referred to as the MPEG1 system or the MPEG2 system) for image information, which has been proposed as a result of the international standardization work by the Oding Expert Group, is called intra-frame coding and inter-frame coding. In combination with performing motion compensation prediction and inter-frame prediction, high-efficiency coding of moving image information is performed. Two-dimensional discrete cosine transform (2
(Dimensional DCT). In the orthogonal transform, a predetermined block size (N.times.M pixels.fwdarw.N pixels.times.vertical M line block size) is calculated for each screen image signal to be subjected to high-efficiency encoding. An image divided for each “unit block” (in the MPEG1 system and the MPEG2 system, a block having a block size of 8 × 8 pixels ← horizontal 8 pixels × vertical 8 lines is referred to as a “unit block”). Performed on signals.

[0006] (N × M) orthogonal transform coefficients (8 × 8 = 6 in the MPEG1 and MPEG2 systems) obtained by orthogonally transforming each block of the unit.
An area of a predetermined size including at least one of the unit blocks described above (the area referred to by the term "macroblock" in the MPEG1 system and the MPEG2 system) includes at least one of the unit blocks described above. That is,
In the MPEG1 system and the MPEG2 system, an area of a block size of 16 × 16 pixels ← 16 pixels × 16 lines for the luminance signal Y and two color difference signals Cr, C
An area composed of a block size area of 8 × 8 pixels ← 8 horizontal pixels × 8 vertical lines for each of b)
It is quantized by the “block quantization width value” set for each. For example, in the MPEG1 system and the MPEG2 system, the “block quantization width value” is indicated as [{macroblock quantization characteristic value (or macroblock quantization scale) QS} × quantization matrix].

[0007] The orthogonal transform coefficient (eg, DCT coefficient) quantized by the block quantization width value is separated into its DC component (DC component) and AC component (AC component). The DC component of the orthogonal transform coefficient (for example, DCT coefficient) is differentially coded, and the AC component of the orthogonal transform coefficient (for example, DCT coefficient) is entropy-coded after zigzag scanning (information based on the bias of the code appearance probability). Amount compression: for example, variable-length encoding such as the Huffman method). The image data transformed and encoded as described above is output as a bit stream (bit string). Next, the decoding operation on the image data transformed and coded as described above is performed in the reverse operation to the above-described coding operation to obtain an output image. When the quantization is performed, quantization noise is generated in the output image due to the unavoidable quantization error. When the complexity of the image to be coded is greater than the transmission rate, the quantization noise significantly degrades the image quality.

In general, among the quantization errors that cause the above-mentioned quantization noise, the quantization error of the low-frequency component is defined as an output image distortion in a state where there is no correlation between unit blocks, that is, a so-called block distortion. Among the quantization errors that occur inside and generate quantization noise, the quantization error of the high frequency component causes ringing-like output image distortion, so-called mosquito noise, around the edge. By the way, as described above, the quantization noise generated in an image is particularly conspicuous in a flat portion of the image, and small noise is added to a place where a large change of the video signal level is caused from a low frequency to a high frequency. In the case of such a quantization noise waveform, noise is hard to be detected because the sensitivity difference in visual characteristics is small. However, when there is a large change in the video signal level only in the low band, and when small noise is added to the high band, the noise is easily detected. As a matter of course,
When a large noise is added, it is needless to say that a fatal error is detected as coding degradation regardless of the low band and the high band. Then, one of the means for solving the problem of image quality deterioration due to the quantization noise as described above.
For example, as disclosed in Japanese Patent Laid-Open No. 4-372074, a solution has been proposed to reduce the quantization noise by applying a post-filter to a decoded image.

[0009]

However, when a means for applying a post-filter to a decoded image, which is disclosed as a means for solving the problem in the above-mentioned Japanese Patent Application Laid-Open No. 4-372074, is applied, only a part of the image in the frame is deteriorated. Even in such a case, since the filter acts uniformly on the entire frame, there is a problem that the image quality of a portion where the image is not degraded deteriorates, and a solution to the problem has been demanded.

[0010]

SUMMARY OF THE INVENTION According to the present invention, an image signal divided into blocks each having a predetermined block size is subjected to orthogonal transformation for each unit block, and then the image signal is divided into at least the aforementioned unit blocks. Is quantized using a block quantization width value that is individually set for each region of a predetermined size including one of the above, and a predetermined region including at least one of the unit blocks described above. When decoding image data that has been transformed and coded by performing inter-frame predictive coding that also uses a motion vector for each size region, it includes one of the above-described unit blocks obtained in the decoding process. In order to perform signal processing such that a high-frequency component is reduced according to a motion vector value of each region of a predetermined size and a difference between motion vector values of adjacent regions. A method for reducing quantization noise that occurs when decoding the transformed and encoded image data, and a decoding apparatus for image data to which the method for reducing quantization noise that occurs when decoding the image data described above, that is, a predetermined method is used. After the image signal divided for each block of the unit having the block size is orthogonally transformed for each block of the unit, the image signal divided for each block of the unit having the predetermined block size is divided into each unit. Is then quantized using at least a block quantization width value that is individually set for each region of a predetermined size including one of the unit blocks described above. And an inter-frame prediction code using a motion vector for each region of a predetermined size including at least one of the unit blocks described above. In order to decode the image data that has been transform-encoded by performing the encoding, at least a buffer memory, a variable-length decoding unit, an inverse quantization unit, an inverse orthogonal transform unit, and an inverse motion compensation unit. ,
An image memory and a decoding apparatus for transform-encoded image data comprising an image memory, wherein a motion vector value individually set for each of the above-described regions and a motion vector value of an adjacent region are obtained. Generate a control signal according to the difference,
By controlling the pass overband characteristic of the variable passband low-pass filtering means capable of changing the passband by the control signal,
An object of the present invention is to provide a decoding apparatus for transform-coded image data which controls the variable passband low-pass filtering means so that the high-frequency component can be adaptively reduced according to the motion vector.

[0011]

A predetermined block size (N × M) is set for the image signal of each screen which is the object of high-efficiency encoding.
For each “unit block” having pixels ← N horizontal pixels × vertical M line block size, orthogonal transformation was performed to obtain (N ×
M) orthogonal transform coefficients are quantized using a “block quantization width value” set for each region of a predetermined size including at least one of the above-described unit blocks, and Image data transformed and coded by performing inter-frame predictive coding also using a motion vector for each region of a predetermined size including at least one of the unit blocks described above, A bit stream based on additional information (for example, block quantization width information, motion vector, prediction mode information, and the like) required when decoding the decoded image data is stored in the buffer memory. In the variable length decoding unit to which the bit stream read from the buffer memory is supplied, the entropy coded (variable length coded) image data and the additional data required when decoding the transform coded image data are added. Information (eg, block quantization width information, motion vector, prediction mode information, etc.) is decoded.

By providing the decoded image data and the block quantization width information in the decoded additional information to the inverse quantization unit, the inverse quantization operation on the image data performed by the inverse quantization unit allows The orthogonal transform coefficient is supplied from the inverse quantization unit to the inverse orthogonal transform unit. In the inverse orthogonal transform unit,
A two-dimensional inverse orthogonal transform is performed for each unit block, and the image data in the frequency domain is inversely transformed into image data in the time domain. The image data in the time domain output from the inverse orthogonal transform unit is added to the image data in the motion compensated state by the motion compensation unit according to a coding type indicating a difference between intra-frame encoding and inter-frame encoding. Or,
Otherwise, the image data is stored in the image memory as output image data.

The motion vector information in the additional information included in the bit stream is detected, and the detected motion vector information, that is, a predetermined size including at least one of the unit blocks is detected. Or a control signal is generated according to the difference between the motion vector value individually set for each region and the motion vector value of an adjacent region, and the high-frequency component is adaptively reduced by the control signal. Such signal processing is performed by controlling the pass band of the variable pass band low-pass filter.

[0014]

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram of a decoding apparatus for transform-coded image data according to the present invention; Will be described in detail. 1 and 2 are block diagrams showing a configuration example of an image data decoding apparatus to which the quantization noise reduction method of the present invention is applied, FIG. 3 is a diagram used for explaining a differential motion vector in an adjacent region, and FIG. Is a block diagram showing a configuration example of a variable passband low-pass filter,
FIG. 5 is a diagram used to explain the state of filtering at the boundary between unit blocks, FIG. 6 is a block diagram showing a configuration example of a control signal generator, and FIG. 7 is an explanatory diagram of a control characteristic example.

In the image data decoding apparatus of the present invention shown in FIGS. 1 and 2, reference numeral 1 denotes an input terminal of a bit stream (bit string) to be decoded. (3, 15 ...)
A decoder integrated circuit) is considered to be an integrated circuit component. In the embodiment of the decoding device shown in each of FIGS. 1 and 2, the components surrounded by the dashed-dotted line frame 3 include at least the buffer memory 8, the variable-length decoding unit 9, A commercially available integrated circuit including an inverse quantization unit 10, an inverse orthogonal transformation unit 11, an addition unit 12, a motion compensation unit 13, and an image memory 14 can be used.

The bit stream supplied to the input terminal 1 is compressed by an orthogonal transformation using an intra-frame (in-frame) correlation (an information amount performed by using a spatial correlation). Compression) and the amount of information compression using the correlation between screens (inter-frames) arranged on the time axis (compression of the amount of information performed using temporal correlation), and the probability of code appearance It is assumed that the image data is image data (for example, image data according to the MPEG1 system and the MPEG2 system) which has been subjected to high-efficiency conversion coding by combining three different compression means for compressing the amount of information due to the bias. In the following description of the present specification, description is made assuming that image data to be decoded is image data according to the MPEG1 system and the MPEG2 system.

The high-efficiency coding of moving picture information in the MPEG1 system and the MPEG2 system is performed by combining intra-frame coding with two-dimensional discrete cosine transform (two-dimensional DCT) and inter-frame coding to perform motion compensation prediction. And inter-frame prediction. Then, the image signal of the screen for each one of the images targeted for the high-efficiency encoding is:
The block is divided into “unit blocks” having a block size of 8 × 8 pixels (8 horizontal pixels × 8 vertical lines), and DCT is performed for each block of each of the above units. Then, the 64 DCT transform coefficients for each block of each unit are quantized by the “block quantization width value”. MPEG1 system,
In the MPEG2 system, the “block quantization width value” is a region called a “macro block” in a region of a predetermined size including one of the unit blocks, that is, a luminance. 16 for signal Y
An area having a block size of × 16 pixels (16 horizontal pixels × 16 vertical lines) and a block size of 8 × 8 pixels (8 horizontal pixels × 8 vertical lines) for each of the two color difference signals Cr and Cb. This is a value indicated by the product of the {macroblock quantization characteristic value (or the quantization scale of the macroblock) QS} set for each area including the size area and the quantization matrix.

The DCT coefficient quantized using the DCT transform coefficient as a dividend and the “block quantization width value” as a divisor has a DC component (DC component) and an AC component (AC component).
Component). The DC component of the DCT coefficient is differentially coded, and the AC component of the DCT coefficient is subjected to zigzag scanning and then entropy coding (information compression by bias of the code appearance probability ... a variable length code such as the Huffman method). The image data that has been transformed and coded is provided with additional information [for example, {block quantization width information →
Macroblock quantization characteristic value (or macroblock quantization scale) QS}, motion vector, prediction mode information, etc.] are added to form a bit stream.
In the image data decoding apparatus of the present invention shown in FIGS. 1 to 6, the bit stream supplied to the input terminal 1 is, for example, a first-in first-out memory (FI
FO) is stored in the buffer memory 8.

The variable-length decoder 9 to which the bit stream read from the buffer memory 8 is supplied is used to decode the entropy-coded (variable-length coded) image data and the transform-coded image data. Necessary additional information, for example, {block quantization width information → macroblock quantization characteristic value (or quantization scale of macroblock) QS}, motion vector, prediction mode information and the like are decoded. The image data decoded by the variable length decoding unit 9 and the block quantization width information {macroblock quantization characteristic value (or quantization scale of macroblock) QS} are supplied to the inverse quantization unit 10. In addition, the motion vector, the prediction mode information, and the like are supplied to the inverse motion compensation unit 13.

The image data decoded by the variable-length decoding unit 9 and the block quantization width information {macroblock quantization characteristic value (or quantization scale of macroblock) QS} in the decoded additional information are Given inverse quantization unit 1
At 0, the DCT transform coefficient obtained by performing the inverse quantization operation is supplied to the inverse orthogonal transform unit (inverse DCT) 11. In an inverse orthogonal transform unit (inverse DCT) 11, two-dimensional inverse DCT is performed for each unit block, image data in the frequency domain is inversely transformed into image data in the time domain, and the resulting data is supplied to the addition unit 12. I do. The image data in the time domain supplied to the addition unit 12 as described above is an image in a state where the motion is compensated by the motion compensation unit 13 according to a coding type indicating a difference between intra-frame coding and inter-frame coding. Whether the data is added to the data or not, the image data is stored in the image memory 14 as output image data. Buffer memory 8, variable length decoding unit 9, inverse quantization unit 10, inverse orthogonal transformation unit 11
, An adder 12, a motion compensator 13, and an image memory 14.
The above description regarding the operation of each of the constituent parts is common to the operation of each of the above-described constituent parts in the image data decoding apparatus of the present invention shown in FIGS.

In the image data decoding apparatus of the present invention shown in FIGS. 1 and 2, the image data from the image memory 14 in the decoder integrated circuit 3 (or 15) is supplied to a variable band low-pass filter. 7 to the output terminal 2. And the variable band low-pass filter 7 is
The control signal supplied from the control signal generator 6 to the variable band low-pass filter 7 controls the signal level of the high-frequency portion of the image signal passing therethrough to change. FIG. 6 shows a specific configuration example of the control signal generator 6 described above. FIG. 4 shows a specific configuration example of the variable band low-pass filter 7 described above. 4 is a component that functions as a variable band low-pass filter in the horizontal direction of the image, and is a component that is indicated by a dotted frame 7v in FIG. The portion is a component that functions as a variable band low-pass filter in the vertical direction of the image.

In the specific configuration of the variable band low-pass filter 7 illustrated in FIG. 4, the variable band low-pass filter functions as a variable band low-pass filter in the horizontal direction of the image indicated by a dotted frame 7h. By connecting the components and the components functioning as the variable band low-pass filter in the vertical direction of the image indicated by the dotted frame 7v in series, the two-dimensional variable band low band as a whole is obtained. It is configured as a pass filter 7. The variable-band low-pass filter 7 is composed of one two-dimensional low-pass filter, a subtractor (a subtractor provided in correspondence with 17 in FIG. 4), and a multiplier (in FIG. 4). A subtractor provided in correspondence with 18) and an adder (a subtractor provided in correspondence with 19 in FIG. 4) may be used.

The variable band low-pass filter 7 illustrated in FIG.
In the above, the component 7h has a horizontal LPF 1 configured to have a predetermined fixed cutoff frequency.
6, a subtractor 17, a multiplier 18, and an adder 19, and the component 7v is
A vertical LPF 20 configured to have a predetermined fixed cutoff frequency, a subtractor 21, a multiplier 22,
And an adder 23. The multiplier 18 (22) provided in the two components 7h and 7v in the variable band low-pass filter 7 shown in FIG. A signal (for example, a coefficient in a range of 0 to 1.0... A multiplication coefficient signal) is supplied. As a result, the output image data from the variable band low-pass filter 7 has a coefficient of 0 to 1.1 with respect to the signal components in the frequency band higher than the cutoff band of the LPF having the predetermined fixed cutoff frequency. The state is multiplied by 0.

That is, a control signal from the control signal generator 6 (for example, a coefficient in the range of 0 to 1.0... A multiplication coefficient signal)
Is a region in which an individual quantization scale value related to a block quantization width used when quantizing DCT transform coefficients is to be set, that is, a unit block (DCT) having a predetermined block size in which DCT is performed. Adaptively to a motion vector value individually set for each region of a predetermined size including at least one of the blocks, and a difference value between each motion vector set to an adjacent region. This is a coefficient for causing the variable band low-pass filter 7 to perform signal processing for reducing high-frequency components.

In each area surrounded by a rectangular frame in FIG. 3, an individual quantization scale value associated with a block quantization width used when quantizing DCT transform coefficients is set. Area, that is, a unit block (D) having a predetermined block size in which DCT is performed.
CT block), and is a region of a predetermined size including at least one of the CT blocks).
V0, MV1, MV2, MV3, MV4, MV5, MV6, M
V7 and MV8 indicate the motion vector values set in the respective areas, and the above-described motion vector MVi
(I is 0, 2, 3,... 8) is represented by the following equation (1).

[0026]

(Equation 1)

In FIG. 3, a region whose motion vector value is indicated as MV0 is defined as a current region, and this current region and eight regions adjacent to the current region (regions having an 8-connected relationship → motion vector If the value is MV1,
MV2, MV3, MV4, MV5, MV6, MV7, and MV8), and the difference value dMV from the motion vectors are the motion vector values MV0, MV1, and MV2 of the nine regions. , MV3, MV4, MV5, MV6,
It is obtained by performing an operation represented by the following equation (2) using MV7 and MV8. MV0, MV1, MV2... MV8
The motion vector of each area indicated by 等 is used for the MV0 using the X coordinate and the Y coordinate indicated in parentheses, and MV (xo, y
o), MV1 is MV (xo-1, yo-1), MV2 is MV (x
o, yo-1), MV3 is MV (xo + 1, yo-1), MV4 is M
V (xo-1, yo) and MV5 are MV (xo + 1, yo) and MV6
Is MV (xo-1, yo + 1) and MV7 is MV (xo, yo +).
1) and MV8 are represented as MV (xo + 1, yo + 1).

[0028]

(Equation 2) Note that MV0 (x) in the equation (2) is MV (xo), MV0
(y) is MV (yo), and MV1 (x) is MV (x
o-1), MV1 (y) is MV (yo-1), MV2 (x) is MV
(Xo), MV2 (y) is MV (yo-1), MV3 (x) is M
V (xo + 1) and MV3 (y) are MV (yo-1) and MV4 (x)
Is MV (xo-1), MV4 (y) is MV (yo), MV5
(x) is MV (xo + 1), MV5 (y) is MV (yo), M
V6 (x) is MV (xo-1), MV4 (y) is MV (yo +
1), MV7 (x) is MV (xo), MV7 (y) is MV (y
o + 1) and MV8 (x) are MV (xo + 1) and MV8 (y) are MV (yo + 1).

The motion vector MVi (MV0 in FIG. 3) and the difference value dMV of the motion vector are standardized between G (x, y) 1.0 and 0.0.
FIG. 7 shows the magnitude of the motion vector MV (or the magnitude of the difference value dMV of the motion vector) on the horizontal axis, and the vertical axis shows the multiplication supplied from the control signal generator 6 to the variable band low-pass filter 7. Setting examples of a plurality of control characteristics A to E for adaptively reducing high-frequency components according to the magnitude of the motion vector MV and the magnitude of the difference vector dMV by taking coefficients (1.0 to 0.0). Is shown. For example, when the magnitude of the difference value dMV of the motion vector is 50 or more, the characteristic is A. When the magnitude of the difference value dMV of the motion vector is 40 or more and less than 50, the B characteristic. When the magnitude of the difference value dMV of the motion vector is 20 or more and less than 30, the D characteristic, and when the magnitude of the difference value of the motion vector dMV is less than 20, the C characteristic when the magnitude is 30 or more and less than 40, , Set the control characteristics.

FIG. 6 shows that a coefficient signal in the range of 0.0 to 1.0 is generated as described above, and that it is output to the variable band low-pass filter 7.
Is a configuration example of the control signal generator 6 (the control signal generator 6 shown in FIGS. 1 and 2) for supplying the control signal, and the control signal generator 6 shown in FIG. The motion vector information of the sequential area is stored in the memory 24 of the motion vector information having a storage capacity capable of storing the motion vector information in the range (for example, at least about three macroblock lines). Memory 2 for motion vector information
4 is read out, and the motion vector information in the adjacent area is compared and determined by the motion vector information in the adjacent areas, for example, by performing the operation of the above-described equation (2) and sequentially calculating the equation (2). The value of the magnitude dMV of the motion vector for the current area is obtained and supplied to the multiplication coefficient setting unit 26.

The multiplication coefficient setting unit 26 uses the value of the magnitude dMV of the motion vector for the sequential current area supplied thereto as an address, and calculates the magnitude of the difference value dMV of the motion vector as illustrated in FIG. And a coefficient signal (control signal) corresponding to a coefficient that satisfies the relationship between the multiplication coefficient (1.0 to 0.0) supplied to the variable band low-pass filter 7 from the control signal generator 6 and the corresponding coefficient signal (control signal) are stored in a ROM table ( Look-up table). The coefficient signal is supplied from the control signal generator 6 to the variable band low-pass filter 7 as a control signal from the control signal transmitter 27. Note that the control signal generator 6 configured to generate a control signal corresponding to the magnitude of the motion vector MV in the sequential area has a configuration in which, when the magnitude of the motion vector MV in the sequential area is given as an address, Those provided with a ROM table (look-up table) for outputting predetermined coefficient signals can be used. Further, the difference value d of the motion vector described above is used as an address to be given to the ROM table (look-up table).
In addition to the size of the MV or the size of the motion vector,
The magnitude of the difference value dMV of the motion vector and the pixel address value, or the magnitude of the motion vector and the pixel address value may be used.

By the way, for each region of a predetermined size including at least one unit block having a predetermined block size in which DCT is performed, a motion vector value set individually or a difference between motion vectors is set. FIG. 5 shows the filtering operation of the low-pass characteristic for the pixels near the boundary of the block of the selected unit by the variable band low-pass filter 7 which performs a signal processing operation for adaptively reducing the high-frequency component with the value. This is done as shown and described. That is, in FIG. 5, t1, t2,
t3... indicate different times in a state shifted by one clock cycle, respectively. FIG. 5 shows a case where the variable band low-pass filter 7 is configured as a low-pass filter having a 5-tap filter length of the FIR filter, near the boundary of a unit block in a pixel array of a specific one row in an image. Pixels p-3, p-2, p-1, p, p + 1, p
+2, p + 3... Are subjected to low-pass filtering.

In the image data decoding apparatus of the present invention shown in FIGS. 1 and 2, the bit stream (bit string) to be decoded supplied to the input terminal 1 is composed of at least the buffer memory 8 and the variable length Image data decoded by a component including a decoding unit 9, an inverse quantization unit 10, an inverse orthogonal transform unit 11, an addition unit 12, a motion compensation unit 13, and an image memory 14. Is image memory 1
4 and can output decoded image data from the image memory 14 as described above. The image data decoding apparatus of the present invention shown in FIGS. 1 and 2 provides the image data read from the image memory 14 to the variable band low-pass filter 7 to The operation of the band low-pass filter 7 causes the image data in a state where the quantization noise is reduced to be output from the output terminal 2 of the image data decoding device.

First, in the image data decoding apparatus of the present invention shown in FIG. 1, the bit stream supplied to the input terminal 1 and to be decoded is supplied to a buffer provided in the decoder integrated circuit 3. The data is supplied to the memory 8 and also supplied to the buffer memory 4. The bit stream read from the buffer memory 4 using a first-in first-out memory is supplied to a motion vector detecting unit 5. As the above-mentioned motion vector detection unit 5, a configuration mode having the same function as the variable length decoding unit 9 provided in the decoder integrated circuit 3 can be used. Then, the above-described motion vector detection unit 5 detects the motion vector information for each of the areas in sequence from the bit stream supplied thereto, and supplies it to the control signal generation unit 6.

Meanwhile, in the image data decoding apparatus of the present invention shown in FIG. 1, the bit stream supplied to the input terminal 1 and to be decoded is provided in a buffer memory provided outside the decoder integrated circuit 3. 4 and supplies the bit stream read out from the buffer memory 4 to a motion vector detecting section 5, which generates a control signal based on the motion vector information for each of the sequential areas detected from the bit stream. Although it is provided to the unit 6, the image data decoding apparatus of the present invention shown in FIG.
By the operation of the variable length decoding unit 9 provided inside the decoder integrated circuit 15, the motion vector information for each of the sequential areas detected from the bit stream is supplied to the control signal generation unit 6.

That is, the image data decoding apparatus of the present invention shown in FIG. 2 is the same as the image data decoding apparatus of the present invention shown in FIG. Long decoding unit 9 and inverse quantization unit 1
0, an inverse orthogonal transform unit 11, an adding unit 12, a motion compensating unit 13, and a buffer memory 4 externally attached to the decoder integrated circuit 3, which is an integrated circuit including an image memory 14. The operation with the motion vector detecting section 5 is performed by utilizing the functions of the buffer memory 8 and the variable length decoding section 9 provided inside the decoder integrated circuit 3, and the variable length decoding section is used. 9, a control signal generator 6 to which the motion vector information for each of the sequential areas detected from the bit stream is given.
Also included in one decoder integrated circuit 15 is a component of variable band low-pass filter 7 which performs a control operation of a high frequency component of image data supplied from image memory 14 according to a control signal output from. In this manner, the integrated circuit is configured.

In the image data decoding apparatus according to the present invention shown in FIGS. 1 and 2, the control signal generator 6 has, for example, the configuration shown in FIG. 6 as described above. In addition, as the variable band low-pass filter 7, for example, one having the configuration as illustrated in FIG. 4 as described above is used. Therefore, in the image data decoding apparatus of the present invention shown in FIGS. 1 and 2, the control signal generated by the control signal generation unit 6, that is, the additional information included in the bit stream as described above, For example, the variable pass band low band is adjusted so that the signal processing for adaptively reducing the high band component is performed based on the magnitude of the motion vector information value MV or the difference value dMV of the motion vector for each of the regions. By controlling the pass band of the pass filter, quantization noise in the image can be satisfactorily reduced.

[0038]

As is apparent from the above description, the method for reducing quantization noise and the apparatus for decoding image data according to the present invention can be applied to each of the images to be encoded with high efficiency. (N × M) image signals obtained by performing orthogonal transformation for each “unit block” having a predetermined block size (N × M pixels ← block size of horizontal N pixels × vertical M lines) for the image signal of the screen Orthogonal transform coefficients are quantized using a “block quantization width value” set for each region of a predetermined size including at least one of the blocks of the unit, and Image data transformed and encoded by performing inter-frame predictive encoding also using a motion vector for each region of a predetermined size including one of the above-described blocks, and the transformed encoded image data The motion vector information in the additional information included in the bit stream based on the additional information (for example, block quantization width information, motion vector, prediction mode information, etc.) required at the time of decoding is detected. According to the motion vector information, that is, a difference between a motion vector value individually set for each region of a predetermined size including at least one unit block and a motion vector value of an adjacent region. In the decoding process, a control signal is generated, and the signal processing for adaptively reducing the high frequency component by the control signal is performed by controlling the pass band of the variable pass band low-pass filter. According to the magnitude of the difference value of the motion vector between the obtained regions,
The size of the high-frequency component is adaptively controlled, and the high-frequency component can be reduced only in the portion where the motion is severe and the quantization noise is generated. In the case where the quantization noise can be removed and the whole image is moving, the motion compensation is relatively easy, and the quantization noise is relatively small. It is better to remove quantization noise while adapting to the property that resolution is easier to recognize when the pattern moves in the same direction as a whole than when each block moves in a different direction. Can be.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration example of an image data decoding device to which a quantization noise reduction method according to the present invention is applied.

FIG. 2 is a block diagram illustrating a configuration example of an image data decoding apparatus to which the quantization noise reduction method according to the present invention is applied.

FIG. 3 is a diagram used to describe a differential motion vector in an adjacent area.

FIG. 4 is a block diagram showing a configuration example of a variable pass band low-pass filter.

FIG. 5 is a diagram used to explain the state of filtering at the boundary between unit blocks.

FIG. 6 is a block diagram illustrating a configuration example of a control signal generator.

FIG. 7 is a characteristic example diagram showing a relationship between a motion vector and a control characteristic.

[Explanation of symbols] [Explanation of symbols]

REFERENCE SIGNS LIST 1 input terminal, 2 output terminal, 3, 15 decoder integrated circuit, 4, 8 buffer memory, 5 motion vector detector, 6 control signal generator, 7 variable band low-pass filter, 9: variable length decoding unit, 10: inverse quantization unit, 11: inverse orthogonal transform unit (inverse DCT unit), 12: addition unit, 13: inverse motion compensation unit, 14: image memory, 16: horizontal LPF, 1
7, 21 ... subtractor, 18, 22 ... multiplier, 19, 23 ...
Adder, 20 ... vertical LPF,

Claims (3)

(57) [Claims] An image signal divided for each unit block having a predetermined block size is subjected to an orthogonal transformation for each unit block, and thereafter, a predetermined signal including at least one of the unit blocks described above is determined. Vector using a block quantization width value individually set for each region of a predetermined size, and a motion vector for each region of a predetermined size including at least one of the unit blocks described above. When decoding image data that has been transformed and coded by performing inter-frame predictive coding using also a predetermined size including one of the above-described unit blocks obtained in the decoding process. Transform coding which performs signal processing to reduce high-frequency components according to a motion vector value of each region and a difference between motion vector values of adjacent regions. A method of reducing quantization noise generated when decoding image data. 2. An image signal divided for each unit block having a predetermined block size is orthogonally transformed for each unit block, and then divided for each unit block having a predetermined block size. After the image signal is orthogonally transformed for each unit block, at least a block quantization width value individually set for each region of a predetermined size including one of the unit blocks described above. And quantized by performing inter-frame predictive coding using a motion vector for each region of a predetermined size including at least one of the unit blocks described above. In order to decode the image data, at least a buffer memory, a variable length decoding unit, an inverse quantization unit, an inverse orthogonal transform unit, an inverse motion compensation unit, An image memory and a decoding device for transform-encoded image data including an image memory, wherein the motion vector detection unit that detects a motion vector individually set for each of the regions, A control signal generating means for generating a control signal according to a difference between a motion vector value individually set for each region sequentially output from the motion vector detection unit and a motion vector value of an adjacent region; A variable pass band low-pass filtering means capable of varying a pass frequency band by a control signal; and the variable pass band low-pass filter adapted to adaptively reduce a high-frequency component according to the control signal output from the control signal generating means. Means for controlling the band-pass filtering means. 3. An image signal divided for each unit block having a predetermined block size is orthogonally transformed for each unit block, and then divided for each unit block having a predetermined block size. After the image signal is orthogonally transformed for each unit block, at least a block quantization width value individually set for each region of a predetermined size including one of the unit blocks described above. And quantized by performing inter-frame predictive coding using a motion vector for each region of a predetermined size including at least one of the unit blocks described above. In order to decode the image data, at least a buffer memory, a variable length decoding unit, an inverse quantization unit, an inverse orthogonal transform unit, an inverse motion compensation unit, An image memory and a decoding apparatus for transform-encoded image data, comprising: a motion vector individually set for each area sequentially output from the variable-length decoding unit. Control signal generating means for generating a control signal in accordance with a difference between a value and a motion vector value of an adjacent area; a variable passband low-pass filtering means capable of changing a pass frequency band by the control signal; and the control signal described above. Means for controlling the variable passband low-pass filtering means so that the high-frequency component can be adaptively reduced in accordance with the control signal output from the generating means. apparatus.